Small Island Developing States (SIDS) are disproportionately exposed to climate-driven disasters, yet often rely on fragile terrestrial networks that fail when they are most needed. TV White Space (TVWS) links offer long-range, low-power coverage; however, current deployments depend on Protocol to Access White Spaces (PAWS) database connectivity for channel authorization, creating a single point of failure during outages. We present SIDSense, an edge AI framework for database-free TVWS operation that preserves regulatory intent through a compliance-gated controller, audit logging, and graceful degradation. SIDSense couples CNN-based spectrum classification with a hybrid sensing-first, authorization-as-soon-as-possible workflow and co-locates sensing and video enhancement with a private 5G stack on a maritime vessel to sustain situational-awareness video backhaul. Field experiments in Barbados demonstrate sustained connectivity during simulated PAWS outages, achieving 94.2% sensing accuracy over 470-698 MHz with 23 ms mean decision latency, while maintaining zero missed 5G Layer-1 (L1) deadlines under GPU-aware scheduling. We release an empirical Caribbean TVWS propagation and occupancy dataset and look to contribute some of the components of the SIDSense pipeline to the open source community to accelerate resilient connectivity deployments in climate-vulnerable regions.

翻译：小岛屿发展中国家（SIDS）在气候灾害面前尤为脆弱，却往往依赖脆弱的地面网络，而这些网络在最需要时常常失效。电视白频谱（TVWS）链路提供了远距离、低功耗的覆盖；然而，当前部署依赖于用于信道授权的“白频谱接入协议”（PAWS）数据库连接，这在网络中断期间构成了单点故障。本文提出SIDSense，一种用于无数据库TVWS运行的边缘人工智能框架，通过合规门控控制器、审计日志记录和优雅降级机制来维护监管意图。SIDSense将基于CNN的频谱分类与“感知优先、授权尽速”的混合工作流相结合，并将频谱感知与视频增强功能同私有5G协议栈共置于海上船只，以维持态势感知视频回传。在巴巴多斯进行的实地实验表明，在模拟PAWS中断期间能够维持连接，在470-698 MHz频段实现了94.2%的感知准确率，平均决策延迟为23毫秒，同时在GPU感知调度下保持了5G层一（L1）零截止期错失。我们发布了一个加勒比地区TVWS传播与占用实证数据集，并计划将SIDSense流程中的部分组件开源，以加速气候脆弱区域的韧性连接部署。